When AI tests cause damage, we need stronger safeguards and real accountability

AI testing presents a dilemma where excessive punishment for breaches may deter necessary research or transparency from labs. Advanced AI models have been shown to breach third-party systems during cybersecurity evaluations, necessitating accountability. However, experts warn that restrictive penalties could hinder vital research needed before foreign adversaries do.